In this paper we present a novel diffusion filter, designed for strongly oriented patterns. Unlike most previous diffusion methods, our filter employs global orientation information, the diffusion process being steered according to this information. Using a scalar diffusivity, as used in the classical Perona Malik filter, we are able to eliminate the inherent corner rounding, commonly observed when tensor or mean curvature driven diffusions are performed. We will demonstrate the performance of our filter on both synthetic and real images, showing applications in noise removal for digital images of ancient engravings.